One of my “favourite external links” is to the weekly market comment written by US economist and fund manager John Hussman. Hussman has been downbeat on the US economy for some time but has argued there was insufficient evidence to forecast a recession. He now thinks the balance has tipped, as explained in last week’s comment, titled “Expecting a recession”.

Other forecasters and media pundits have also been piling on the gloom recently. The Economist  this week opined that "recession in America looks increasingly likely".

Hussman's approach is admirably empirical. He describes a “rule of thumb” based on four conditions that have been jointly observed in every US recession. The conditions are:

1. A widening of credit spreads from six months earlier.

2. A flat yield curve, defined as longer-term Treasury yields no more than 2.5% above three-month yields.

3. A fall in the stock market from six months earlier.

4. A purchasing managers’ index for manufacturing of 54 or lower coupled with either non-farm employment growth of less than 1.3% over the prior 12 months or a rise of 0.4 percentage points or more in the unemployment rate from its 12-month low.

Conditions 1, 2 and 4 were met in October and condition 3 is likely to fall into place in November – the S&P 500 has averaged 1478 month-to-date compared with 1511 in May. Hussman therefore now believes a recession is immediately ahead.

The economy was much stronger than the bears forecast in the second and third quarters. I have been expecting a sharp slowdown in growth in the fourth quarter but no recession, at least yet. Should I change my view in light of Hussman’s analysis?

I have to concede that his rule of thumb works well historically. There have been eight US recessions since 1950, according to the National Bureau of Economic Research. Hussman’s indicator gives a signal either before or during all eight. Even more impressively, there are no false signals.

However, it bothers me that the indicator ignores information on the magnitude of the underlying variables. One might reasonably expect the values of the change in credit spreads, yield curve slope, change in stock prices etc. to be relevant to the assessment of the probability of a recession.

To investigate this, I estimated a statistical model for assessing whether the economy is currently in a recession using the values of the Hussman variables. I included current and six-month-ago values to allow for lags in the relationship. The fitted probability estimates of the model are shown in the chart below. Historical performance is similar to the rule of thumb, with all eight recessions since 1950 signalled by the probability rising above 50% and no false signals. However, unlike the simple rule, the model has yet to flash red in the current cycle, with a latest reading of 20%.

I described my own recession probability indicators in an earlier post. The version including credit spreads has been rising recently but has also yet to breach 50% (current reading 45%).

Downside economic risks have clearly increased with further weakness in credit markets and rises in energy costs but I still think a recession can be avoided.